DE102016009203B4 - Injection molding system for calculating optimal operating conditions and machine learning device therefor - Google Patents

Abstract

Injection molding system (1), which is provided with at least one injection molding machine (2) and has an artificial intelligence that carries out machine learning, the injection molding system (1) comprising: a status observation section (21), which, when an injection molding from the injection molding machine (2 ) is performed, physical amounts related to the injection molding under execution; a physical quantity data storing section (22) that stores the physical quantity data observed by the status observation section (21); a judgment condition setting section (23) of the judgment conditions for machine learning; a judgment calculation section (24) that calculates a judgment based on the physical quantity data observed by the status observation section (21) and the judgment conditions set by the judgment condition setting section (23); an operating condition adjustment learning section nitt (25) that performs machine learning to adjust operating conditions based on the judgment calculated by the judgment calculation section (24) and the operating conditions set in the injection molding system (1) and the data on physical quantities; a learning result storage section (26) that learns a machine learning result by the operating condition adjustment learning section (25); and an operating condition adjustment height output section (27) that determines and outputs an operating condition to be adjusted and an adjustment height based on the learning result by the operating condition adjustment learning section (25).

Description

BACKGROUND OF THE INVENTION

Field of the Invention

The present invention relates to an injection molding system and in particular to an injection molding system with which the calculation of optimal operating conditions is implemented without adaptation by a user, and to a corresponding machine learning device.

Description of the Related Art

When manufacturing a molded article that is used to mold new molded articles, it is necessary to calculate the optimal values of operating conditions, including molding conditions, before mass production of the molded articles based on the molded article begins. In a condition setting operation that calculates the optimal operating conditions of an injection molding machine, a user is required to adjust various operating conditions to obtain the optimal operating conditions by referring to process monitoring data or the measurement of weights of molded articles and confirming a molding status by visually recognizing a molded article while the operating conditions are set based on his experience as rough standards and a molding operation is performed. For this reason, it is necessary for a user to take time to calculate optimal operating conditions by adjusting various operating conditions and comparing molded articles that have been formed on the basis of the various operating conditions.

Meanwhile, as conventional technologies for supporting the above operation of setting molding conditions by a user, a technology in which molding conditions or molding data is previously stored in a non-volatile memory and displayed for comparison purposes, a technology in which past molding conditions in response to a request from one User read and used, and the like (see, e.g., Japanese Patent Application Laid-Open No. JP H06- 39 889 A and Japanese Patent Application Laid-Open No. JP H11-333 899 A ).

In the above operating condition setting operation by a user, calculating optimal operating conditions depending on the skill of the user who is performing the operation takes time or there is a difference in the level (quality) of the optimal operating conditions depending on the user. For this reason, it is difficult to calculate operating conditions under the same standard every time.

In addition, in the above operating condition setting operation, it is important to derive operating conditions that implement a reduction in energy consumption during molding in view of the manufacturing cost of molded articles in a mass production. However, it is difficult even for a trained user to derive operating conditions for molding with reduced energy consumption while maintaining the high quality of the molded products.

As in Japanese Patent Application Laid-Open No. JP H06- 39 889 A and in Japanese Patent Application Laid-Open No. JP H11-333 899 A disclosed, such problems may not be solved by merely saving previous shaping conditions and using them.

The WO 2014/150 632 A1 describes an injection molding system with self-learning capabilities, in which a controller learns over time how certain changes affect the actual results of the injection molded parts, so that adjustments to the heating elements of the hot runner can be made independently.

The WO 02/047 884 A1 describes a visual inspection of a molded part integrated in the injection unit.

The EP 2 400 358 A1 describes a self-learning system that can independently adjust manufacturing parameters to improve a manufacturing result.

SUMMARY OF THE INVENTION

In view of the above circumstances, an object of the present invention is to provide an injection molding system with which it is possible to adjust operating conditions including molding conditions in a short period of time and to carry out molding with less energy consumption.

This object is achieved according to the invention by an injection molding system with the features of claim 1 and a machine learning device with the features of claim 17.

An embodiment of the present invention provides an injection molding system with the features of claim 1, which is provided with at least one injection molding machine and has an artificial intelligence that performs machine learning, the injection molding system including: a status observation section that, when injection molding is performed by the injection molding machine, observes physical amounts for injection molding while performing; a physical quantity data storing section that stores the physical quantity data observed by the status observation section; a judgment condition setting section that sets judgment conditions for machine learning; a judgment calculation section that calculates a reward based on the physical quantity data observed by the status observation section and the judgment conditions set by the judgment condition setting section; an operating condition adjustment learning section that performs machine learning to adjust operating conditions based on the judgment calculated by the judgment calculation section and the operating conditions set in the injection molding system and the physical quantity data; a learning result storage section that stores a learning result of machine learning by the operating condition adjustment learning section; and an operating condition adjustment height output section that determines and outputs an operating condition to be adjusted and an adjustment height based on the learning result by the operating condition adjustment learning section.

In the injection molding system, the learning result stored in the learning result storage section can be used in teaching the operating condition adjustment learning section.

The injection molding system can also include a measuring section. In the injection molding system, the data on physical quantities observed by the status observation section can be at least one of a weight and a size of a molded article as measured by the measuring section, an appearance, a length, an angle, an area and a volume as calculated from image data for the molded article, of an optical inspection result of an optical molded article and a measurement result of a strength of the molded article, and the physical quantity data storing section may also store physical quantity data other than the physical quantity data of the molded article.

In the injection molding system, input of the judgment conditions into the judgment condition setting section can be implemented by a display device provided in the injection molding machine.

In the injection molding system, if at least one of stabilizing data on physical amounts, reducing cycle time and saving energy is achieved, the evaluation calculation section can output a positive evaluation according to an achievement level.

In the injection molding system, when at least one event from destabilization of physical quantity data, an increase in cycle time and an increase in energy consumption occurs, the judgment calculation section can output a negative judgment according to a degree of event.

In the injection molding system, an allowable value can be set in advance in the physical amount data, and the judgment calculation section can give a positive judgment if the physical amount data falls within the allowable value.

In the injection molding system, an allowable value can be set in advance in the physical quantity data, and if the physical quantity data deviate from the allowable value, the judgment calculation section can give a negative judgment based on a deviation amount.

In the injection molding system, a target value can be set in advance in the physical quantity data, and when the physical quantity data is close to the target value, the judgment calculation section can give a positive judgment based on a discrepancy between the target value and the physical quantity data.

In the injection molding system, a target value can be set in advance in the physical quantity data, and if the physical quantity data deviates from the target value, the judgment calculation section can give a negative judgment based on a difference in amount between the target value and the physical quantity data.

In the injection molding system, when a status indicating a molding defect occurs, the judgment calculation section can output a negative judgment according to a molding defect degree.

In the injection molding system, the molding error includes at least one of a ridge, a sink mark, a curvature, an air bubble, a short shot, a flow line, a welding line, a silver streak, a color irregularity, a discoloration, a carbonization, a penetration of impurities, a deviation from one optical axis of a lens molded article of an allowable value and an error of a thickness of a molded article.

In the injection molding system, the operating conditions subjected to the machine learning by the operating condition adjustment learning section include at least one of molding clamping conditions, pusher conditions, spray holding conditions, metering conditions, temperature conditions, nozzle contact conditions, resin supply conditions, molding thickness conditions, molded article removal conditions, and hot runner conditions.

The injection molding system may further include a robot that serves as a molded article removal unit in which the molded article removal conditions are set.

In the injection molding system, at least one of the operating conditions can be varied within a prescribed range to be learned from the operating condition adjustment learning section.

In the injection molding system, each of the plurality of injection molding systems may have a communication section for communicating with an external environment, and physical quantity data stored in each of the physical quantity data storage sections and a learning result stored in each of the Learning result storage sections are stored, sent / received to be shared.

Another embodiment of the present invention provides a machine learning device having the features of claim 17. A machine learning device not according to the present invention is disclosed that performs machine learning to adjust operating conditions by the injection molding machine, the machine learning device including: a learning result storage section that stores a learning result of adjusting the operating conditions; a status observation section that, when an injection molding is performed by the injection molding machine, observes data on physical quantities for injection molding under execution; and an operating condition adjustment height output section that determines and outputs an operating condition to be adjusted and an adjustment height based on the learning result stored in the learning result storage section.

According to one embodiment of the present invention, it becomes possible to adapt various operating conditions in a short period of time and to carry out a more stable shaping by using machine teaching in the calculation of optimal operating conditions. It also makes it possible to carry out molding with less energy consumption. Furthermore, it becomes possible to implement the machine learning, with which a better result can be obtained for each of the injection molding systems, in such a way that shaping data to be adapted or learning data for machine learning are shared and used in the plurality of injection molding systems.

Figure list

• 1 ein Schaubild zum Beschreiben des Grundkonzepts eines Verstärkungslernalgorithmus ist;
• 2 ein schematisches Konfigurationsschaubild ist, das ein Spritzgusssystem gemäß einer Ausführungsform der vorliegenden Erfindung zeigt; und
• 3 ein Schaubild ist, das ein Beispiel für das Anzeigen von Spritzhaltedruckdaten bei einem Schuss unter Verwendung einer Druckwellenform zeigt.

The above and other objects and features of the present invention will become apparent from the descriptions of the following embodiments with reference to the accompanying drawings, in which:

• 1 Fig. 3 is a diagram for describing the basic concept of a gain learning algorithm;
• 2nd FIG. 12 is a schematic configuration diagram showing an injection molding system according to an embodiment of the present invention; and
• 3rd FIG. 12 is a graph showing an example of displaying shot hold pressure data on a shot using a pressure waveform.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

An embodiment of the present invention will be described below with reference to the drawings.

In the present invention, a machine learning device that serves as artificial intelligence is inserted into an injection molding system to perform machine learning on injection molding operating conditions, thereby adjusting the various injection molding operating conditions. This is an injection molding system with which it is possible to calculate optimal operating conditions in a short period of time in order to further stabilize the molding and to achieve further energy savings. (1) Machine teach-in

In general, the machine teach-in according to its goal or conditions in various algorithms such. B. classified learning and unsupervised learning. An object of the present invention is to learn the operation of setting operating conditions for a molding. Taking into account the fact that the above injection molding system has parameters or the like that are not directly measurable under a spraying environment, and is difficult in this, To explicitly indicate which type of action (adaptation of operating conditions) is to be suitably carried out in relation to a molded article resulting from an injection, an amplification learning algorithm is used in which a machine learning device automatically learns an action to achieve a goal only if an assessment is accepted.

1. (1) Der Akteur beobachtet einen Umgebungsstatus s_t zu einem gewissen Zeitpunkt.
2. (2) Der Akteur wählt eine Aktion a_t , die er durchführen darf, auf Basis eines Beobachtungsergebnisses und vergangenem Lernen aus und führt diese durch.
3. (3) Der Umgebungsstatus s_t verändert sich zu einem nächsten Status s_t+1 , nachdem die Aktion a_t durchgeführt wurde.
4. (4) Der Akteur akzeptiert eine Beurteilung r_t+1 auf Basis der Statusänderung infolge der Aktion a_t .
5. (5) Der Akteur entwickelt das Einlernen auf Basis des Status s_t , der Aktion a_t , der Beurteilung r_t+1 und eines bisherigen Lernergebnisses weiter.

1 Fig. 4 is a diagram for describing the basic concept of a gain learning algorithm. In reinforcement learning, a learning actor and an action are further developed through the interactions between an actor (machine learning device), which acts as a learning subject, and a movement (control target system), which serves as a tax target. The following interactions between the actor and the environment are carried out in more detail.

1. (1) The actor observes an environmental status s _t at some point.
2. (2) The actor chooses an action a _t which he is allowed to carry out on the basis of an observation result and past learning and carries them out.
3. (3) The environmental status s _t changes to a next status s _{t + 1} after the action a _t was carried out.
4. (4) The actor accepts an assessment r _{t + 1} based on the status change following the action a _t .
5. (5) The actor develops the learning based on the status s _t , the action a _t , judging r _{t + 1} and a previous learning outcome.

When teaching the above processing ( 5 ) the actor records the assignment of an observed status s _t , an action a _t and an assessment r _{t + 1} as reference information for determining a quantity of an assessment that he may receive in the future. For example, if the number of statuses that the actor is allowed to have is m each time and the number of actions that the actor is allowed to perform is n, the actor receives a two-dimensional arrangement of m × n, with assessments r _{t + 1} , the pairs of status s _t and actions a _t can be saved by repeatedly performing actions.

In the case of a value function (evaluation function), which indicates to what degree a current status or an action is useful, based on the above-recorded assignment, the actor then updates the value function (evaluation function) while the actions are carried out repeatedly by one learn optional action that corresponds to a status.

A status value function is a value function that shows the degree to which a certain status s _t is useful. The status value function is expressed as a function using a status as an argument and repeated based on a judgment obtained with respect to an action in a certain status, a value of a future status changed with the action, or the like when learning from Actions updated. The update formula of the status value function is defined according to a gain learning algorithm. For example, when learning with temporal difference (TD), which is displayed as one of the gain learning algorithms, the status value function is defined by the following formula (1). Note that in the following formula (1) α indicates a learning coefficient, γ indicates a discount rate, α and γ are defined as 0 <α ≤ 1 or 0 <γ ≤ 1. $$V\left(s_t\right)\leftarrow V\left(s_t\right)\text{+}\alpha \left[{\text{+}}_{}\text{+}\gamma \left({\text{+}}_{}\right)\left({}_{}\right)\right]$$

In addition, an action value function is a value function that shows the degree to which an action a _t in a certain status s _t is useful. The action value function is expressed as a function using a status and an action as arguments and based on a judgment obtained with respect to an action in a certain status, an action value of a future status that is changed with the action, or the like Learn from repeated actions updated. The update formula of the action value function is defined according to a gain learning algorithm. For example, in Q-Learning, which is displayed as one of the typical reinforcement learning algorithms, the action value function is defined by the following formula (2). Note that in the following formula (2) α indicates a learning coefficient, γ indicates a discount rate, α and γ are defined as 0 <α ≤ 1 or 0 <γ ≤ 1. $$\left({}_{}{}_{}\right)\leftarrow \left({}_{}{}_{}\right)\text{+}\alpha \left({}_{}\text{+}\underset{a}{\gamma max}\left({\text{+}}_{}\right)-\left({}_{}{}_{}\right)\right)$$

Note that as a method for storing a value function (evaluation function), a supervised learning device such as B. a support vector machine (SVM) and a neural network of a multi-value output, the one value (evaluation) with a status s _t and an action a _t outputs as inputs, e.g. B. when the status s can take many statuses or the like can be used in addition to a method using an approximation function and a method using an arrangement.

Furthermore, when choosing an action in the above processing ( 2nd ) an action a _t with which an assessment (r _{t + 1} + r _{t + 2} + ...) for a future in a current status s _t reached its maximum value (an action to change to a most useful status when a status value function is used, or a most useful action in status when an action value function is used), selected using a value function (evaluation function) generated by previous learning. Note that while learning, an actor can select an arbitrary action with constant probability to support learning when choosing an action in the above processing ( 2nd ) to develop further (epsilon (ε) greedy method).

As described above, the learning is accomplished by repeatedly performing the above processing operations ( 1 ) to ( 5 ) further developed. Even in a new environment after completing learning in a certain environment, learning can be developed to adapt to the new environment through additional learning. Accordingly, when the above processing is applied to the operation of setting operating conditions as in the present invention, the additional learning to operate the setting of operating conditions where the molded part of the reformed article is set as a new environment is based on the previous learning to operate the setting of the operating conditions, which makes it possible to adjust various conditions in a short period of time.

In addition, reinforcement learning uses a system in which a plurality of actors are connected via a network or the like, and information on status s, actions a, judgments r or the like is shared among the actors and applied to each learning, whereby each of the Actors perform a scattered reinforcement learning taking into account the surroundings of the other actors in order to be able to carry out efficient learning. In the present embodiment, when a plurality of actors (machine learning devices) controlling a plurality of environments (injection molding machines as control targets) perform scattered reinforcement learning in a state in which they are connected via a network or the like, the actors are also in able to efficiently learn the operation of setting operating conditions for a molding.

Note that even if various methods such as e.g. B. Q-learning, a SARSA method, TD learning and an AC method commonly known as gain learning algorithms, any of the above gain learning algorithms can be applied to the present invention. Note that since each of the gain learning algorithms is well known, the detailed description thereof is omitted in the document.

The following is a description of the injection molding system according to the present invention based on a specific embodiment in which a machine learning device is incorporated.

Embodiment

2nd FIG. 12 is a diagram showing the schematic configuration of an injection molding system according to an embodiment of the present invention. An injection molding system 1 the embodiment is by an injection molding machine 2nd , a molded part 3rd , Tax targets such as B. other peripheral devices, a machine learning device 20th which acts as artificial intelligence, which carries out machine learning, or the like. If the in 2nd shown configuration with the elements of the im 1 reinforcement learning shown is compared, the machine learning device corresponds 20th the actor and the whole, including the injection molding machine 2nd , the molding 3rd and tax targets such as B. other peripherals, corresponds to the environment.

The injection molding system 1 is formed by various devices, each of which can be provided with a control device, a sensor or the like.

The control device of the injection molding machine 2nd includes a device for controlling the temperature of a nozzle, a cylinder and the lower part of a funnel, a device for controlling the pressure, a device for controlling a nozzle touch force or the like, in addition to a device for controlling the part clamping and a part-related device 7 such as a mold clamping force control device and a device 9 for controlling the opening / closing of a molded part, the adjustment of the thickness of a molded part, the ejector, the rotation of a screw, the back and forth movement of the screw, the back and forth movement of an injection unit and a device for controlling an injection unit-related device 10th such as a rotation of a supplier for a corresponding quantity feeding device or the like.

The molding control device 3rd includes a device for controlling a temperature of a molded article, a device for controlling a hot runner temperature and opening / closing a hot runner nozzle, a device for controlling a movable member for injection molding, a device for controlling a core set and a core pull, a device for controlling a sliding table and a turntable for multicolor molding or the like, a device for switching a flow path of a material for multicolor molding, a device for oscillating a molding or the like.

The peripheral devices include a shaped article removal device (robot) 5 by a device 8th for controlling the robot, a device for inserting an insert, a device for feeding a film for in-mold treatment, a device for feeding a jacket ring for jacket ring molding, a device for filling gas for gas-assisted molding, a device for filling of gas for molded foam using a supercritical fluid, a device for filling long fibers, a device for mixing two components together for LIM molding, a device for removing burrs from a molded article, a device for cutting a sprue, a scale for a molded article, a machine for examining strength of the molded article, an apparatus for optically examining a molded article, an apparatus for imaging the molded article and an image processing device, a robot for conveying a molded article or the like.

Some of the above control devices are provided with a sensor to perform feedback control or feed forward control based on a closed control loop. In addition, some of the above control devices are only used to output data.

Furthermore, the machine learning device 20th that performs machine learning with a status observation section 21st , a section 22 for storing data on physical quantities, a judgment condition setting section 23 , an assessment calculation section 24th , an operating condition adjustment session 25th , a learning result storage section 26 and an operating condition adjustment height output section 27th Mistake. The machine learning device 20th can be inside the injection molding machine 2nd can be provided or can be in a personal computer or the like outside the injection molding machine 2nd be provided.

The status monitoring section 21st is a functional section that observes physical quantity data for injection molding from each of the devices of the injection molding system 1 are output, and imports the observed status data into the machine learning device 20th . The physical quantity data includes a temperature, a position, a speed, an acceleration, a current, a voltage, a pressure, a time, image data, image analysis data, a torque, an energy, a distortion, an energy consumption, a mold opening height, a Return flow height, a tie rod deformation height, a heater heating ratio, a weight of a molded article, a strength of a molded article, a size of a molded article, an appearance calculated from image data for a molded article, a length of each part of a molded article, an angle, an area, a volume, an optical Examination result of an optically shaped article, a measurement result of a strength of a molded article, a height of deviation of an optical axis of a transparent molded article, a calculation value calculated by the arithmetic processing of each physical quantity, or the like. Combinations of the above elements of data on physical quantities define the environmental status s used for machine learning.

The section 22 for storing physical amount data is a functional section that receives and stores physical amount data and the stored physical amount data to the judgment calculation section 24th and the operating condition adjustment session 25th issues. The section 22 to store physical amount data stores physical amount data by the status observation section 21st observed during injection molding as data on physical amounts of an injection molded article. Input of physical quantity data into the section 22 for storing data on physical quantities can be data that was acquired by the last molding company or data that was acquired by a past molding company. It is also possible that the section 2nd for storing physical quantity data Receives and stores physical quantity data in other injection molding systems 1 or a centralized management system 30th are saved, or is it possible that the section 22 to store data on physical quantities Data on physical quantities described in section 22 to store data on physical quantities stored on other injection molding systems 1 or a centralized management system 30th issues.

The judgment condition setting section 23 is a functional section that sets conditions for assigning a machine learning assessment. Positive and negative assessments are given and can be set accordingly. In addition, an input to the judgment condition setting section 23 be carried out via a personal computer, a tablet terminal or the like in the centralized management system 30th is used. By input via a display device 6 the injection molding machine 2nd however, it becomes possible to make simpler settings.

The assessment calculation section 24th analyzes input of physical quantity data from the status observation section 21st or the section 22 for storing data on physical amounts based on conditions set by the judgment condition setting section 23 can be set, and gives a calculated judgment to the operating condition adjustment session 25th out. The judgment output from the judgment calculation section 24th corresponds to an assessment r which is used in machine learning.

The following is an example of judgment conditions by the judgment condition setting section 23 can be set according to the embodiment.

(Assessment 1: A case in which a positive assessment is made when at least one is obtained from stabilizing data on physical amounts, reducing cycle time and saving energy)

When determining the stabilization of physical quantity data, if there is a reduction in the fluctuations due to statistical processing of the physical quantity data, a positive assessment can be made according to the degree of reduction thereof. A standard deviation is generally used as the index for the fluctuations.

When determining the reduction of a cycle time, if the cycle time is reduced, a positive judgment is given according to its degree of reduction.

In energy saving, when the energy consumption of the injection molding machine alone, the energy consumption of the entire injection molding system, the total energy consumption of a plurality of injection molding systems or the like is reduced as an index, a positive judgment is given according to its degree of reduction.

In contrast, when the destabilization of physical quantity data, the prolongation of a cycle time and an increase in energy consumption are caused, a negative judgment is given according to their degrees. (Assessment 2: A case where allowable values are set in advance in physical quantity data and the judgment calculation section gives a positive judgment when the physical quantity data is within allowable values)

If it is known that burrs occur in a molded article at a maximum injection pressure of 200 MPa or more and that a short-circuit occurs in the molded article at a maximum injection pressure of 190 MPa or less, an injection pressure in one injection step is caused to cause a maximum pressure of 200 MPa and a minimum pressure of 190 MPa as permissible values. If data on physical quantities are within the permissible values, a positive assessment is given. On the other hand, if the data on physical quantities are not within the permissible values, a negative assessment can be made according to the amount of deviation. This means that the greater the deviation, the greater the negative assessment to be awarded.

As in 3rd , it is also known that the screen of the injection molding machine has the function of displaying injection holding pressure data in one shot using a pressure waveform based on a horizontal axis (time or screw position) and a vertical axis (pressure), and the function of displaying a warning message or deterioration if the waveform is not within upper and lower limits set in a plurality of intervals of the pressure waveform. It is also possible to make a judgment on the pressure waveform, with the upper and lower limit values set for the pressure waveform from the judgment condition setting section 23 can be set as permissible values.

(Assessment 3: A case where a target value is pre-set in physical quantity data and the judgment calculation section gives a positive judgment according to a difference between the target quantity and physical quantity data when the physical quantity data is close to the target value)

For a target value that is set for a weight of a molded article based on the molded part design and the resin selection, a larger positive judgment can be given when the weight of the molded article comes closer to the target value.

In contrast, a negative judgment based on the amount of variation between the target value and the physical quantity data can be given if the physical quantity data deviates from the target value. Provided that another negative judgment based on a change ratio is given when the change ratio of the amount of deviation increases, it becomes possible to give a much larger negative judgment when the Deviation height increases with increasing speed.

By combining the above setting of allowable values in assessment 2nd and the above setting of a target value in judgment 3rd a target value close to an upper permissible limit value can also be set. A resin used in injection molding has factors that cause variations in molecular weight distribution and melt time viscosity depending on the batch, even if from the same class. For this reason, the likelihood of a short shot occurring near a lower allowable limit is likely to increase even if molding is performed within the allowable values. To avoid this risk, the target value is set to be less than or equal to the upper allowable limit and is set to be close to the upper allowable limit. Thus, the formation is stabilized within the range of allowable values and close to the upper allowable limit, making it possible to reduce the likelihood of a short shot occurring. As described above, it is also possible to use a plurality of judgment conditions in combination.

(Assessment 4: A case in which a negative assessment is given when a status occurs indicating a molding error)

A negative assessment is given if a design error such as e.g. B. a ridge, a sink mark, a curvature, an air bubble, a short shot, a flow line, a weld line, a silver streak, color irregularities, discoloration, carbonization, penetration of contaminants, a deviation of an optical axis of a lens molded article outside a permissible value and an error in the thickness of a molded article occurs in an image obtained by photographing the molded article or in image analysis data obtained by analyzing the image or by an optical inspection device or the like.

In addition, a size of a negative judgment can be changed according to a degree of such an error. For example, when discoloration occurs, the degree of discoloration is digitized by a colorimeter, an image analysis of a photographed image or the like, and a size of a negative judgment is changed according to the degree of discoloration.

Referring again to 2nd conducts the operating condition adjustment session 25th machine learning (reinforcement learning) based on data on physical quantities, the adjustment of operating conditions, including molding conditions of the injection molding system as performed by itself, and one from the judgment calculation section 24th calculated assessment by. At this time, the operating condition adjustment session 25th machine learning (reinforcement learning) using one in the learning result storage section 26 perform saved learning result, which will be described later.
After that, the operating condition adjustment height output section gives 27th , which will be described later, based on a learning result of the operating condition adjustment learning section 25th Adjustment levels of operating conditions for the injection molding system such. B. mold clamping conditions, pusher conditions, spray holding conditions, metering conditions, temperature conditions, nozzle contact conditions, resin supply conditions, mold thickness conditions, molded article removal conditions, hot runner conditions and setting conditions of the control device or the like of each of the peripheral devices. Here, adapting the operating conditions corresponds to action a, which is used when teaching the machine.

Here is the one from the operating condition adjustment session 25th carried out machine teaching a status s _t defined by the combination of data on physical quantities at a certain point in time t, and the adjustment of operating conditions for the injection molding system in relation to the defined status s _t is performed, and the output of an adjustment result from the operating condition adjustment height output section 27th which will be described later correspond to an action a _t . Thereafter, a value corresponds to that of the judgment calculation section 24th is calculated based on data obtained as a result of an injection molding according to the adjustment result r _{t + 1} . A value function that is used in learning is determined in accordance with an applied learning algorithm. For example, if Q-Learning is applied, it is only necessary to have an action value function Q (s _t , a _t ) update according to formula (2) above to further develop learning.

Provided that each of the operating conditions is set to its initial value in advance, at least one of the operating conditions can be varied within a prescribed range to perform the learning. For example, the learning is performed based on a physical amount obtained when a screw rotation speed upon dosing from the initial value of 100 rpm is automatically increased by 10 rpm to perform molding and a physical amount that is obtained when a back pressure from the initial value of 5 MPa is automatically increased by 1 MPa to carry out the molding. Thus, it becomes possible to learn the combination of a screw rotation speed and a back pressure at which the energy consumption reaches its minimum value within a range in which an excellent article can be molded stably without a molding defect.

In addition, it may be possible to use an ε-greedy method as described above and to select an arbitrary action with a prescribed probability in order to further develop the learning.

Note that the device 5 for removing a molded article for which the molded article removal conditions are set, can be an articulated robot. Since a general removal machine in metal injection molding (MIM injection molding) or the like must not remove an extremely fragile molded article (blank), a robot can remove the molded article at low speed with minimal holding force. However, since a molded article removal speed has a larger effect on a cycle time, it is preferable to learn how to remove a molded article at a higher speed so that it is not broken. It is possible to output and determine the presence or absence of the breaking of a molded article as physical amount data based on measurement of a weight, image analysis or the like of the molded article.

The learning outcome storage section 26 stores a learning result of the operating condition adjustment learning section 25th . When a learning result is again from the operating condition adjustment session 25th is used, there is also the learning result storage section 26 a stored learning result to the operating condition adjustment session 25th out. As described above, a learning result can be stored such that a value function corresponding to a machine learning algorithm to be used in a supervised learning device such as B. an SVM (support vector machine) and a neural network of an approximation function, an arrangement or a multi-value output or the like is stored.

Note that it is also possible that the learning result storage section 26 receives and stores a learning result that in other injection molding systems 1 or the centralized system 30th is stored, or it is possible that the learning result storage section 26 one in the learning result storage section 26 saved learning result to the injection molding systems 1 or the centralized management system 30th issues.

The operating condition adjustment height output section 27th determines an operation condition to be adjusted and an adjustment amount based on a learning result of the operation conditions by the operation condition adjustment session 25th be learned and output them. For example, when determining an adjustment amount, the combinations of operating conditions to be adjusted and adjustment amounts that may have been made in a current status become according to an evaluation function based on the learning of the operating condition adjustment learning section 25th evaluated and among these the combination with the highest evaluation is determined. After outputting an adjustment amount from the operating condition adjustment amount output section 27th the learning is then repeated using data on physical amounts that are re-entered into the status observation section 21st can be entered. This makes it possible to get a better learning outcome. In addition, an arbitrary action with a prescribed probability, as described above, can be selected to further develop the learning.

It is also possible to perform machine learning using physical quantity data and an evaluation function in which an operation is expressed by an argument to obtain a maximum judgment. Machine enrollment can be performed while physical quantity data is being collected for the last formation, or can be performed using collected physical quantity data stored in the section on storing physical quantity data.

When performing the machine learning, it is possible to vary at least one of the operating conditions within a prescribed range that of the operating condition adjustment learning section 25th is to learn. By specifically varying the operating conditions to be learned, it becomes possible to learn the influence on the fluctuation efficiently.

If any of the plurality of injection molding systems 1 further provided with a section that is used to communicate with an external environment, it also becomes possible to store data on physical amounts contained in the sections 22 for storing data on physical amounts, and a learning result stored in each of the learning result storage sections 26 is saved to send / receive and share. This enables more efficient machine teaching. For example, when operating conditions are varied within a prescribed range for performing the learning, the learning can be performed efficiently, so that data on physical quantities and learning data among the injection molding systems 1 be exchanged to further develop learning in parallel while varying operating conditions within a prescribed range to be molded in the plurality of injection molding systems 1 perform.

To provide physical quantity and learning data under a variety of injection molding systems 1 , as described above, communication via a host computer such. B. the centralized management system 30th injection molding systems can be carried out 1 communicate directly with each other or a cloud can be used. To process large amounts of data, however, a communication section with a faster communication speed is preferably provided.

The embodiment of the present invention has been described above. However, the present invention is not limited only to the example of the above embodiment, and can be carried out in various aspects with appropriate modifications.

Claims (17)

Injection molding system (1), which is provided with at least one injection molding machine (2) and has an artificial intelligence that carries out machine learning, the injection molding system (1) comprising: a status observation section (21) which, when an injection molding is performed by the injection molding machine (2), observes physical amounts related to the injection molding under execution; a physical quantity data storage section (22) which stores the physical quantity data observed by the status observation section (21); a judgment condition setting section (23) that sets judgment conditions for machine learning; a judgment calculation section (24) that calculates a judgment based on the physical quantity data observed by the status observation section (21) and the judgment conditions set by the judgment condition setting section (23); an operating condition adjustment learning section (25) that performs machine learning to adjust operating conditions based on the judgment calculated by the judgment calculating section (24) and the operating conditions set in the injection molding system (1) and the physical quantity data; a learning result storage section (26) that stores a learning result of machine learning by the operating condition adjustment learning section (25); and an operating condition adjustment height output section (27) that determines and outputs an operating condition to be adjusted and an adjustment height based on the learning result by the operating condition adjustment learning section (25). Injection molding system (1) Claim 1 wherein the learning result stored in the learning result storage section (26) is used in learning the operating condition adjustment learning section (25). Injection molding system (1) Claim 1 or 2nd , further comprising: a measurement section, the data observed by the status observation section (21) on physical amounts of at least one of a weight and a size of a molded article as measured by the measurement section, an appearance, a length, an angle, an area and a volume as calculated from image data for the molded article, an optical examination result of an optical molded article and a measurement result of a strength of the molded article, and the section for storing data on physical quantities (22) also includes other physical quantity data such as the physical quantity data to the molded article stores. Injection molding system (1) according to one of the Claims 1 to 3rd wherein an input of the judgment conditions into the judgment condition setting section (23) can be implemented by a display device (6) provided in the injection molding machine (2). Injection molding system (1) according to one of the Claims 1 to 4th wherein when at least one of stabilizing data on physical amounts, reducing cycle time and energy saving is achieved, the judgment calculation section (24) outputs a positive judgment according to a degree of achievement. Injection molding system (1) according to one of the Claims 1 to 5 wherein when at least one event from destabilization of data on physical amounts, an extension of a cycle time and an increase in energy consumption occurs, the judgment calculation section (24) outputs a negative judgment according to an event degree. Injection molding system (1) according to one of the Claims 1 to 6 wherein an allowable value is set in advance in the physical amount data, and the judgment calculation section (24) outputs a positive judgment when the physical amount data is within the allowable value. Injection molding system (1) according to one of the Claims 1 to 7 wherein a permissible value is set in advance in the physical quantity data, and the judgment calculation section (24), when the physical quantity data deviates from the permissible value, outputs a negative judgment based on a variance amount. Injection molding system (1) according to one of the Claims 1 to 8th wherein a target value is set in advance in the physical amount data, and the judgment calculating section (24), when the physical amount data is close to the target value, outputs a positive judgment based on a discrepancy between the target value and the physical amount data. Injection molding system (1) according to one of the Claims 1 to 9 wherein a target value is previously set in the physical amount data, and the judgment calculating section (24), when the physical amount data deviates from the target value, outputs a negative judgment based on a difference in amount between the target value and the physical amount data. Injection molding system (1) according to one of the Claims 1 to 10th wherein when a status occurs indicating a molding error, the judgment calculation section (24) outputs a negative judgment according to a molding error degree. Injection molding system (1) Claim 11 , wherein the forming error at least one of a ridge, a sink, a curvature, an air bubble, a short shot, a flow line, a welding line, a silver streak, a color irregularity, a discoloration, a carbonization, a penetration of impurities, a deviation of an optical axis of a lens molded article of an allowable value and an error of a thickness of a molded article. Injection molding system (1) according to one of the Claims 1 to 12 wherein the operating conditions subjected to the machine learning by the operating condition adjustment learning section (25) include at least one of molding clamping conditions, pusher conditions, spray holding conditions, metering conditions, temperature conditions, nozzle contact conditions, resin supply conditions, molding thickness conditions, molded article removal conditions and hot runner conditions. Injection molding system (1) Claim 13 , which further comprises: a robot (5) which serves as a shaped article removal unit in which the shaped article removal conditions are set. Injection molding system (1) according to one of the Claims 1 to 14 wherein at least one of the operating conditions is varied within a prescribed range to be learned by the operating condition adjustment learning section (25). Injection molding system (1) according to one of the Claims 1 to 15 , wherein each of a plurality of injection molding systems has a communication section for communicating with an external environment and physical quantity data stored in each of the physical quantity data storage section (22) and a learning result stored in each of the learning result storage sections (26) is stored, sent and / or received for common use. A machine learning apparatus comprising: an injection molding machine (2) performed when an injection molding from the status observation section (21) being observed physical amounts related to the injection molding under execution; a physical quantity data storage section (22) which stores the physical quantity data observed by the status observation section (21); a judgment condition setting section (23) that sets judgment conditions for machine learning; a judgment calculation section (24) that calculates a judgment based on the physical quantity data observed by the status observation section (21) and the judgment conditions set by the judgment condition setting section (23); an operating condition adjustment learning section (25) that performs machine learning to adjust operating conditions based on the judgment calculated by the judgment calculating section (24) and the operating conditions set in the injection molding system (1) and the physical quantity data; a learning result storage section (26) that stores a learning result of machine learning by the operating condition adjustment learning section (25); and an operating condition adjustment height output section (27) that determines and outputs an operating condition to be adjusted and an adjustment height based on the learning result by means of the operating condition adjustment learning section (25).

Images